Gene Networks Inference through Linear Grouping of Variables

The inference of gene networks from gene expression data is an open problem due to the large dimensionality (number of genes) and the small number of data samples typically available, even considering the fact that the network is sparse (limited number of input genes per target gene). In this work we propose a method that alleviates the curse of dimensionality by grouping predictor gene configurations in their respective linear combination values. Each linear combination value results in an equivalence class. In this way, the number of configurations of predictor values becomes a linear function of the dimensionality (number of predictors) instead of an exponential function when considering the original configurations. The proposed method follows the probabilistic gene networks approach which applies local feature selection to obtain an adequate predictor gene set for each gene. Even considering that some information from the original configurations of predictors is lost after applying the grouping, the results indicate that the inference with linear grouping tends to provide networks with better topological similarities than those obtained without grouping in cases where the number of samples is quite limited and the inference involves a larger number of predictors per gene.